| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /*********************************************************************************/ | ||
| 2 | /* Copyright 2009-2026 Barcelona Supercomputing Center */ | ||
| 3 | /* */ | ||
| 4 | /* This file is part of the DLB library. */ | ||
| 5 | /* */ | ||
| 6 | /* DLB is free software: you can redistribute it and/or modify */ | ||
| 7 | /* it under the terms of the GNU Lesser General Public License as published by */ | ||
| 8 | /* the Free Software Foundation, either version 3 of the License, or */ | ||
| 9 | /* (at your option) any later version. */ | ||
| 10 | /* */ | ||
| 11 | /* DLB is distributed in the hope that it will be useful, */ | ||
| 12 | /* but WITHOUT ANY WARRANTY; without even the implied warranty of */ | ||
| 13 | /* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the */ | ||
| 14 | /* GNU Lesser General Public License for more details. */ | ||
| 15 | /* */ | ||
| 16 | /* You should have received a copy of the GNU Lesser General Public License */ | ||
| 17 | /* along with DLB. If not, see <https://www.gnu.org/licenses/>. */ | ||
| 18 | /*********************************************************************************/ | ||
| 19 | |||
| 20 | #ifdef HAVE_CONFIG_H | ||
| 21 | #include <config.h> | ||
| 22 | #endif | ||
| 23 | |||
| 24 | #include "LB_comm/shmem_barrier.h" | ||
| 25 | |||
| 26 | #include "LB_core/DLB_kernel.h" | ||
| 27 | #include "LB_comm/shmem.h" | ||
| 28 | #include "apis/dlb_errors.h" | ||
| 29 | #include "support/atomic.h" | ||
| 30 | #include "support/debug.h" | ||
| 31 | #include "support/mask_utils.h" | ||
| 32 | #include "support/mytime.h" | ||
| 33 | #include "support/types.h" | ||
| 34 | #include "talp/talp.h" | ||
| 35 | |||
| 36 | #include <errno.h> | ||
| 37 | #include <pthread.h> | ||
| 38 | #include <stdio.h> | ||
| 39 | #include <string.h> | ||
| 40 | #include <stdlib.h> | ||
| 41 | |||
| 42 | typedef struct barrier_flags { | ||
| 43 | bool initialized:1; | ||
| 44 | bool lewi:1; | ||
| 45 | } barrier_flags_t; | ||
| 46 | |||
| 47 | typedef struct barrier_t { | ||
| 48 | char name[BARRIER_NAME_MAX]; | ||
| 49 | barrier_flags_t flags; | ||
| 50 | unsigned int participants; | ||
| 51 | atomic_uint ntimes; | ||
| 52 | atomic_uint count; | ||
| 53 | pthread_barrier_t barrier; | ||
| 54 | pthread_rwlock_t rwlock; | ||
| 55 | } barrier_t; | ||
| 56 | |||
| 57 | typedef struct { | ||
| 58 | bool initialized; | ||
| 59 | int max_barriers; // capacity | ||
| 60 | int num_barriers; // size, although detached may be counted | ||
| 61 | barrier_t barriers[]; | ||
| 62 | } shdata_t; | ||
| 63 | |||
| 64 | enum { SHMEM_BARRIER_VERSION = 7 }; | ||
| 65 | enum { SHMEM_TIMEOUT_SECONDS = 1 }; | ||
| 66 | |||
| 67 | static int max_barriers = 0; | ||
| 68 | static shmem_handler_t *shm_handler = NULL; | ||
| 69 | static shdata_t *shdata = NULL; | ||
| 70 | static const char *shmem_name = "barrier"; | ||
| 71 | |||
| 72 | 1 | static void cleanup_shmem(void *shdata_ptr, int pid) { | |
| 73 | |||
| 74 | 1 | bool shmem_empty = true; | |
| 75 | 1 | shdata_t *shared_data = shdata_ptr; | |
| 76 | 1 | int num_barriers = shared_data->num_barriers; | |
| 77 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | for (int i = 0; i < num_barriers; i++) { |
| 78 | 1 | barrier_t *barrier = &shared_data->barriers[i]; | |
| 79 |
1/2✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
|
1 | if (--barrier->participants == 0) { |
| 80 | 1 | *barrier = (const barrier_t){}; | |
| 81 | } else { | ||
| 82 | ✗ | shmem_empty = false; | |
| 83 | } | ||
| 84 | } | ||
| 85 | |||
| 86 | /* If there are no registered barriers, make sure shmem is reset */ | ||
| 87 |
1/2✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
|
1 | if (shmem_empty) { |
| 88 | 1 | memset(shared_data, 0, shmem_barrier__size()); | |
| 89 | } | ||
| 90 | 1 | } | |
| 91 | |||
| 92 | 116 | static void open_shmem(const char *shmem_key, int shmem_size_multiplier) { | |
| 93 | |||
| 94 | 116 | max_barriers = mu_get_system_size() * shmem_size_multiplier; | |
| 95 | 232 | shm_handler = shmem_init((void**)&shdata, | |
| 96 | 116 | &(const shmem_props_t) { | |
| 97 | 116 | .size = shmem_barrier__size(), | |
| 98 | .name = shmem_name, | ||
| 99 | .key = shmem_key, | ||
| 100 | .version = SHMEM_BARRIER_VERSION, | ||
| 101 | .cleanup_fn = cleanup_shmem, | ||
| 102 | }); | ||
| 103 | 116 | } | |
| 104 | |||
| 105 | 145 | int shmem_barrier__init(const char *shmem_key, int shmem_size_multiplier) { | |
| 106 | |||
| 107 |
2/2✓ Branch 0 taken 30 times.
✓ Branch 1 taken 115 times.
|
145 | if (shm_handler != NULL) return DLB_ERR_NOMEM; |
| 108 | |||
| 109 | 115 | open_shmem(shmem_key, shmem_size_multiplier); | |
| 110 | |||
| 111 | 115 | int error = DLB_SUCCESS; | |
| 112 | 115 | shmem_lock(shm_handler); | |
| 113 | { | ||
| 114 | // Initialize some values if this is the 1st process attached to the shmem | ||
| 115 |
2/2✓ Branch 0 taken 80 times.
✓ Branch 1 taken 35 times.
|
115 | if (!shdata->initialized) { |
| 116 | 80 | shdata->initialized = true; | |
| 117 | 80 | shdata->num_barriers = 0; | |
| 118 | 80 | shdata->max_barriers = max_barriers; | |
| 119 | } else { | ||
| 120 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 35 times.
|
35 | if (shdata->max_barriers != max_barriers) { |
| 121 | ✗ | error = DLB_ERR_INIT; | |
| 122 | } | ||
| 123 | } | ||
| 124 | } | ||
| 125 | 115 | shmem_unlock(shm_handler); | |
| 126 | |||
| 127 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 115 times.
|
115 | if (error == DLB_ERR_INIT) { |
| 128 | ✗ | warning("Cannot attach to Barrier shmem because existing size differ." | |
| 129 | " Existing shmem size: %d, expected: %d." | ||
| 130 | " Check for DLB_ARGS consistency among processes or clean up shared memory.", | ||
| 131 | ✗ | shdata->max_barriers, max_barriers); | |
| 132 | } | ||
| 133 | |||
| 134 |
1/2✓ Branch 0 taken 115 times.
✗ Branch 1 not taken.
|
115 | if (error == DLB_SUCCESS) { |
| 135 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 115 times.
|
115 | verbose(VB_BARRIER, "Barrier Module initialized."); |
| 136 | } | ||
| 137 | |||
| 138 | 115 | return error; | |
| 139 | } | ||
| 140 | |||
| 141 | 1 | void shmem_barrier_ext__init(const char *shmem_key, int shmem_size_multiplier) { | |
| 142 | |||
| 143 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | if (shm_handler != NULL) return; |
| 144 | |||
| 145 | 1 | open_shmem(shmem_key, shmem_size_multiplier); | |
| 146 | } | ||
| 147 | |||
| 148 | 145 | void shmem_barrier__finalize(const char *shmem_key, int shmem_size_multiplier) { | |
| 149 |
2/2✓ Branch 0 taken 31 times.
✓ Branch 1 taken 114 times.
|
145 | if (shm_handler == NULL) { |
| 150 | /* barrier_finalize may be called to finalize existing process | ||
| 151 | * even if the file descriptor is not opened. (DLB_PreInit + forc-exec case) */ | ||
| 152 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 31 times.
|
31 | if (shmem_exists(shmem_name, shmem_key)) { |
| 153 | ✗ | shmem_barrier_ext__init(shmem_key, shmem_size_multiplier); | |
| 154 | } else { | ||
| 155 | 31 | return; | |
| 156 | } | ||
| 157 | } | ||
| 158 | |||
| 159 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 114 times.
|
114 | verbose(VB_BARRIER, "Finalizing Barrier Module"); |
| 160 | |||
| 161 | 114 | shmem_finalize(shm_handler, NULL /* do not check if empty */); | |
| 162 | 114 | shm_handler = NULL; | |
| 163 | } | ||
| 164 | |||
| 165 | 1 | int shmem_barrier_ext__finalize(void) { | |
| 166 | // Protect double finalization | ||
| 167 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | if (shm_handler == NULL) { |
| 168 | ✗ | return DLB_ERR_NOSHMEM; | |
| 169 | } | ||
| 170 | |||
| 171 | // Shared memory destruction | ||
| 172 | 1 | shmem_finalize(shm_handler, NULL /* do not check if empty */); | |
| 173 | 1 | shm_handler = NULL; | |
| 174 | 1 | shdata = NULL; | |
| 175 | |||
| 176 | 1 | return DLB_SUCCESS; | |
| 177 | } | ||
| 178 | |||
| 179 | 122 | int shmem_barrier__get_max_barriers(void) { | |
| 180 | 122 | return max_barriers; | |
| 181 | } | ||
| 182 | |||
| 183 | // Basically for testing purposes, num_barriers is not decremented if a barrier | ||
| 184 | // in the middle is detached, so we count the number of initialized barriers. | ||
| 185 | 6 | int shmem_barrier__get_num_barriers(void) { | |
| 186 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
|
6 | if (shdata == NULL) return -1; |
| 187 | |||
| 188 | 6 | int real_num_barriers = 0; | |
| 189 | 6 | shmem_lock(shm_handler); | |
| 190 | { | ||
| 191 | 6 | int num_barriers = shdata->num_barriers; | |
| 192 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 6 times.
|
16 | for (int i = 0; i < num_barriers; ++i) { |
| 193 |
1/2✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 | if (shdata->barriers[i].flags.initialized) { |
| 194 | 10 | ++real_num_barriers; | |
| 195 | } | ||
| 196 | } | ||
| 197 | } | ||
| 198 | 6 | shmem_unlock(shm_handler); | |
| 199 | |||
| 200 | 6 | return real_num_barriers; | |
| 201 | } | ||
| 202 | |||
| 203 | /* Given a barrier_name, find whether the barrier is registered in the shared | ||
| 204 | * memory. Note that this function is not thread-safe */ | ||
| 205 | 10 | barrier_t* shmem_barrier__find(const char *barrier_name) { | |
| 206 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 9 times.
|
10 | if (shm_handler == NULL) return NULL; |
| 207 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 6 times.
|
9 | if (barrier_name == NULL) return NULL; |
| 208 | |||
| 209 | 6 | int num_barriers = shdata->num_barriers; | |
| 210 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 3 times.
|
8 | for (int i = 0; i < num_barriers; ++i) { |
| 211 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | if (shdata->barriers[i].participants > 0 |
| 212 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | && shdata->barriers[i].flags.initialized |
| 213 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 2 times.
|
5 | && strncmp(shdata->barriers[i].name, barrier_name, |
| 214 | BARRIER_NAME_MAX-1) == 0) { | ||
| 215 | 3 | return &shdata->barriers[i]; | |
| 216 | } | ||
| 217 | } | ||
| 218 | |||
| 219 | 3 | return NULL; | |
| 220 | } | ||
| 221 | |||
| 222 | /* Register and attach process to barrier. | ||
| 223 | * barrier_name: barrier name | ||
| 224 | * lewi: whether this barrier does lewi | ||
| 225 | */ | ||
| 226 | 139 | barrier_t* shmem_barrier__register(const char *barrier_name, bool lewi) { | |
| 227 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 139 times.
|
139 | if (shm_handler == NULL) return NULL; |
| 228 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 138 times.
|
139 | if (barrier_name == NULL) return NULL; |
| 229 | |||
| 230 | /* Obtain the shared memory lock to find the appropriate place for the new | ||
| 231 | * barrier. If the barrier is not created, the first process must | ||
| 232 | * initialize it before releasing the lock. If the barrier was already | ||
| 233 | * created, new processes may attach after acquiring both the shmem and the | ||
| 234 | * barrier specific lock. */ | ||
| 235 | 138 | int participants = -1; | |
| 236 | 138 | barrier_t *barrier = NULL; | |
| 237 | 138 | shmem_lock(shm_handler); | |
| 238 | { | ||
| 239 | /* Find a new spot, or an already registered barrier */ | ||
| 240 | 138 | barrier_t *empty_spot = NULL; | |
| 241 | 138 | int num_barriers = shdata->num_barriers; | |
| 242 |
2/2✓ Branch 0 taken 72 times.
✓ Branch 1 taken 93 times.
|
165 | for (int i = 0; i < num_barriers; ++i) { |
| 243 |
1/2✓ Branch 0 taken 72 times.
✗ Branch 1 not taken.
|
72 | if (empty_spot == NULL |
| 244 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 72 times.
|
72 | && shdata->barriers[i].participants == 0 |
| 245 | ✗ | && !shdata->barriers[i].flags.initialized) { | |
| 246 | ✗ | empty_spot = &shdata->barriers[i]; | |
| 247 | } | ||
| 248 |
1/2✓ Branch 0 taken 72 times.
✗ Branch 1 not taken.
|
72 | else if (shdata->barriers[i].participants > 0 |
| 249 |
1/2✓ Branch 0 taken 72 times.
✗ Branch 1 not taken.
|
72 | && shdata->barriers[i].flags.initialized |
| 250 |
2/2✓ Branch 0 taken 45 times.
✓ Branch 1 taken 27 times.
|
72 | && strncmp(shdata->barriers[i].name, barrier_name, |
| 251 | BARRIER_NAME_MAX-1) == 0) { | ||
| 252 | 45 | barrier = &shdata->barriers[i]; | |
| 253 | 45 | break; | |
| 254 | } | ||
| 255 | } | ||
| 256 | |||
| 257 | /* If barrier nor empty_spot are found, try to get a new position from the end */ | ||
| 258 |
5/6✓ Branch 0 taken 93 times.
✓ Branch 1 taken 45 times.
✓ Branch 2 taken 93 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 91 times.
✓ Branch 5 taken 2 times.
|
138 | if (barrier == NULL && empty_spot == NULL && num_barriers < max_barriers) { |
| 259 | 91 | empty_spot = &shdata->barriers[num_barriers]; | |
| 260 | 91 | ++shdata->num_barriers; | |
| 261 | } | ||
| 262 | |||
| 263 | /* New barrier, lock first and initialize required fields */ | ||
| 264 |
4/4✓ Branch 0 taken 93 times.
✓ Branch 1 taken 45 times.
✓ Branch 2 taken 91 times.
✓ Branch 3 taken 2 times.
|
229 | if (barrier == NULL && empty_spot != NULL) { |
| 265 | 91 | barrier = empty_spot; | |
| 266 | 91 | *barrier = (const barrier_t){}; | |
| 267 | |||
| 268 | pthread_rwlockattr_t rwlockattr; | ||
| 269 | 91 | pthread_rwlockattr_init(&rwlockattr); | |
| 270 | 91 | pthread_rwlockattr_setpshared(&rwlockattr, PTHREAD_PROCESS_SHARED); | |
| 271 | 91 | pthread_rwlock_init(&barrier->rwlock, &rwlockattr); | |
| 272 | 91 | pthread_rwlockattr_destroy(&rwlockattr); | |
| 273 | |||
| 274 | /* Initialize only once the lock is acquired */ | ||
| 275 | 91 | pthread_rwlock_wrlock(&barrier->rwlock); | |
| 276 | { | ||
| 277 | 91 | barrier->flags = (const barrier_flags_t) { | |
| 278 | .initialized = true, | ||
| 279 | .lewi = lewi, | ||
| 280 | }; | ||
| 281 | 91 | barrier->participants = 1; | |
| 282 | 91 | participants = 1; | |
| 283 | 91 | snprintf(barrier->name, BARRIER_NAME_MAX, "%s", barrier_name); | |
| 284 | pthread_barrierattr_t barrierattr; | ||
| 285 | 91 | pthread_barrierattr_init(&barrierattr); | |
| 286 | 91 | pthread_barrierattr_setpshared(&barrierattr, PTHREAD_PROCESS_SHARED); | |
| 287 | 91 | pthread_barrier_init(&barrier->barrier, &barrierattr, barrier->participants); | |
| 288 | 91 | pthread_barrierattr_destroy(&barrierattr); | |
| 289 | } | ||
| 290 | 91 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 291 | |||
| 292 |
2/2✓ Branch 0 taken 45 times.
✓ Branch 1 taken 2 times.
|
47 | } else if (barrier != NULL){ |
| 293 | /* Barrier was created by another participant, attach. | ||
| 294 | * (timedlock is used to avoid potential deadlocks) */ | ||
| 295 | struct timespec timeout; | ||
| 296 | 45 | get_time_real(&timeout); | |
| 297 | 45 | timeout.tv_sec += SHMEM_TIMEOUT_SECONDS; | |
| 298 | 45 | int timedwrlock_error = pthread_rwlock_timedwrlock(&barrier->rwlock, &timeout); | |
| 299 |
1/2✓ Branch 0 taken 45 times.
✗ Branch 1 not taken.
|
45 | if (likely(timedwrlock_error == 0)) { |
| 300 | /* Update participants */ | ||
| 301 | 45 | participants = ++barrier->participants; | |
| 302 | |||
| 303 | /* Create new barrier with the number of participants updated */ | ||
| 304 | 45 | pthread_barrier_destroy(&barrier->barrier); | |
| 305 | pthread_barrierattr_t attr; | ||
| 306 | 45 | pthread_barrierattr_init(&attr); | |
| 307 | 45 | pthread_barrierattr_setpshared(&attr, PTHREAD_PROCESS_SHARED); | |
| 308 | 45 | pthread_barrier_init(&barrier->barrier, &attr, barrier->participants); | |
| 309 | 45 | pthread_barrierattr_destroy(&attr); | |
| 310 | |||
| 311 | 45 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 312 | |||
| 313 | ✗ | } else if (timedwrlock_error == ETIMEDOUT || timedwrlock_error == EDEADLK) { | |
| 314 | ✗ | shmem_unlock(shm_handler); | |
| 315 | ✗ | fatal("Timed out while creating shmem_barrier.\n" | |
| 316 | "Please, report at " PACKAGE_BUGREPORT); | ||
| 317 | ✗ | } else if (timedwrlock_error == EINVAL) { | |
| 318 | ✗ | shmem_unlock(shm_handler); | |
| 319 | ✗ | fatal("Error acquiring timedwrlock while creating shmem_barrier.\n" | |
| 320 | "Please, report at " PACKAGE_BUGREPORT); | ||
| 321 | } | ||
| 322 | } | ||
| 323 | } | ||
| 324 | 138 | shmem_unlock(shm_handler); | |
| 325 | |||
| 326 |
2/2✓ Branch 0 taken 136 times.
✓ Branch 1 taken 2 times.
|
138 | if (barrier != NULL) { |
| 327 | /* Attach if needed and print number of participants */ | ||
| 328 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 136 times.
|
136 | verbose(VB_BARRIER, "Attached to barrier %s. Participants: %d", |
| 329 | barrier->name, participants); | ||
| 330 | } | ||
| 331 | |||
| 332 | 138 | return barrier; | |
| 333 | } | ||
| 334 | |||
| 335 | /* The attach function should not have any races with the global shared memory. | ||
| 336 | * At most, 'barrier' points to an unitialized barrier and error is returned. | ||
| 337 | * */ | ||
| 338 | 8 | int shmem_barrier__attach(barrier_t *barrier) { | |
| 339 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 | if (barrier == NULL) return DLB_ERR_UNKNOWN; |
| 340 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 | if (shm_handler == NULL) return DLB_ERR_NOSHMEM; |
| 341 | |||
| 342 | #ifdef DEBUG_VERSION | ||
| 343 | int count; | ||
| 344 | #endif | ||
| 345 | int participants; | ||
| 346 | 8 | pthread_rwlock_wrlock(&barrier->rwlock); | |
| 347 | { | ||
| 348 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 1 times.
|
8 | if (!barrier->flags.initialized |
| 349 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 7 times.
|
7 | || barrier->participants == 0) { |
| 350 | 1 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 351 | 1 | return DLB_ERR_PERM; | |
| 352 | } | ||
| 353 | |||
| 354 | /* Update participants */ | ||
| 355 | 7 | participants = ++barrier->participants; | |
| 356 | |||
| 357 | /* Create new barrier with the number of participants updated */ | ||
| 358 | 7 | pthread_barrier_destroy(&barrier->barrier); | |
| 359 | pthread_barrierattr_t attr; | ||
| 360 | 7 | pthread_barrierattr_init(&attr); | |
| 361 | 7 | pthread_barrierattr_setpshared(&attr, PTHREAD_PROCESS_SHARED); | |
| 362 | 7 | pthread_barrier_init(&barrier->barrier, &attr, barrier->participants); | |
| 363 | 7 | pthread_barrierattr_destroy(&attr); | |
| 364 | |||
| 365 | #ifdef DEBUG_VERSION | ||
| 366 | 7 | count = barrier->count; | |
| 367 | #endif | ||
| 368 | } | ||
| 369 | 7 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 370 | |||
| 371 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 7 times.
|
7 | ensure(count == 0, "Barrier Shared memory inconsistency while attaching " |
| 372 | "barrier (count = %d).\n" "Please, report at " PACKAGE_BUGREPORT, | ||
| 373 | count); | ||
| 374 | |||
| 375 | 7 | return participants; | |
| 376 | } | ||
| 377 | |||
| 378 | /* The detach function may remove the barrier if 'participants' reaches 0 and | ||
| 379 | * compete with a barrier creation. This function needs to acquire both locks. */ | ||
| 380 | 1010 | int shmem_barrier__detach(barrier_t *barrier) { | |
| 381 |
2/2✓ Branch 0 taken 868 times.
✓ Branch 1 taken 142 times.
|
1010 | if (barrier == NULL) return DLB_ERR_UNKNOWN; |
| 382 |
2/2✓ Branch 0 taken 39 times.
✓ Branch 1 taken 103 times.
|
142 | if (shm_handler == NULL) return DLB_ERR_NOSHMEM; |
| 383 | |||
| 384 | #ifdef DEBUG_VERSION | ||
| 385 | 103 | int count = -1; | |
| 386 | #endif | ||
| 387 | 103 | int participants = -1; | |
| 388 | |||
| 389 | 103 | shmem_lock(shm_handler); | |
| 390 | { | ||
| 391 | struct timespec timeout; | ||
| 392 | 103 | get_time_real(&timeout); | |
| 393 | 103 | timeout.tv_sec += SHMEM_TIMEOUT_SECONDS; | |
| 394 | 103 | int timedwrlock_error = pthread_rwlock_timedwrlock(&barrier->rwlock, &timeout); | |
| 395 |
1/2✓ Branch 0 taken 103 times.
✗ Branch 1 not taken.
|
103 | if (likely(timedwrlock_error == 0)) { |
| 396 | |||
| 397 |
2/2✓ Branch 0 taken 102 times.
✓ Branch 1 taken 1 times.
|
103 | if (!barrier->flags.initialized |
| 398 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 102 times.
|
102 | || barrier->participants == 0) { |
| 399 | 1 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 400 | 1 | shmem_unlock(shm_handler); | |
| 401 | 1 | return DLB_ERR_PERM; | |
| 402 | } | ||
| 403 | |||
| 404 | #ifdef DEBUG_VERSION | ||
| 405 | 102 | count = barrier->count; | |
| 406 | #endif | ||
| 407 | /* Both locks are acquired and barrier is valid, detach: */ | ||
| 408 | |||
| 409 | /* Update participants */ | ||
| 410 | 102 | participants = --barrier->participants; | |
| 411 | |||
| 412 |
2/2✓ Branch 0 taken 43 times.
✓ Branch 1 taken 59 times.
|
102 | if (participants > 0) { |
| 413 | /* Create new barrier with the number of participants updated */ | ||
| 414 | 43 | pthread_barrier_destroy(&barrier->barrier); | |
| 415 | pthread_barrierattr_t attr; | ||
| 416 | 43 | pthread_barrierattr_init(&attr); | |
| 417 | 43 | pthread_barrierattr_setpshared(&attr, PTHREAD_PROCESS_SHARED); | |
| 418 | 43 | pthread_barrier_init(&barrier->barrier, &attr, barrier->participants); | |
| 419 | 43 | pthread_barrierattr_destroy(&attr); | |
| 420 | |||
| 421 | 43 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 422 | } else { | ||
| 423 | /* If this is the last participant, uninitialize barrier */ | ||
| 424 | 59 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 425 | 59 | pthread_barrier_destroy(&barrier->barrier); | |
| 426 | 59 | pthread_rwlock_destroy(&barrier->rwlock); | |
| 427 | 59 | *barrier = (const barrier_t){}; | |
| 428 | |||
| 429 | /* Try to compact barrier list */ | ||
| 430 |
2/2✓ Branch 0 taken 65 times.
✓ Branch 1 taken 53 times.
|
118 | for (int i = shdata->num_barriers - 1; i >= 0; --i) { |
| 431 |
2/2✓ Branch 0 taken 59 times.
✓ Branch 1 taken 6 times.
|
65 | if (!shdata->barriers[i].flags.initialized) { |
| 432 | 59 | --shdata->num_barriers; | |
| 433 | } else { | ||
| 434 | 6 | break; | |
| 435 | } | ||
| 436 | } | ||
| 437 | } | ||
| 438 | ✗ | } else if (timedwrlock_error == ETIMEDOUT || timedwrlock_error == EDEADLK) { | |
| 439 | ✗ | shmem_unlock(shm_handler); | |
| 440 | ✗ | fatal("Timed out while detaching barrier.\n" | |
| 441 | "Please, report at " PACKAGE_BUGREPORT); | ||
| 442 | ✗ | } else if (timedwrlock_error == EINVAL) { | |
| 443 | ✗ | shmem_unlock(shm_handler); | |
| 444 | ✗ | fatal("Error acquiring timedwrlock while detaching barrier.\n" | |
| 445 | "Please, report at " PACKAGE_BUGREPORT); | ||
| 446 | } | ||
| 447 | } | ||
| 448 | 102 | shmem_unlock(shm_handler); | |
| 449 | |||
| 450 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 102 times.
|
102 | ensure(count == 0, "Barrier Shared memory inconsistency while attaching " |
| 451 | "barrier (count = %d).\n" "Please, report at " PACKAGE_BUGREPORT, | ||
| 452 | count); | ||
| 453 | |||
| 454 | 102 | return participants; | |
| 455 | } | ||
| 456 | |||
| 457 | 48 | void shmem_barrier__barrier(barrier_t *barrier) { | |
| 458 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 48 times.
|
48 | if (unlikely(shm_handler == NULL)) return; |
| 459 | |||
| 460 | 48 | pthread_rwlock_rdlock(&barrier->rwlock); | |
| 461 | { | ||
| 462 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 47 times.
|
48 | if (unlikely(!barrier->flags.initialized)) { |
| 463 | 1 | warning("Trying to use a non initialized barrier"); | |
| 464 | 1 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 465 | 1 | return; | |
| 466 | } | ||
| 467 | |||
| 468 | 47 | unsigned int participant_number = DLB_ATOMIC_ADD_FETCH(&barrier->count, 1); | |
| 469 | 47 | bool last_in = participant_number == barrier->participants; | |
| 470 | |||
| 471 |
1/4✗ Branch 0 not taken.
✓ Branch 1 taken 47 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
47 | verbose(VB_BARRIER, "Entering barrier %s%s", barrier->name, last_in ? " (last)" : ""); |
| 472 | |||
| 473 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 15 times.
|
47 | if (last_in) { |
| 474 | // Barrier | ||
| 475 | 32 | pthread_barrier_wait(&barrier->barrier); | |
| 476 | |||
| 477 | // Increase ntimes counter | ||
| 478 | 32 | DLB_ATOMIC_ADD_RLX(&barrier->ntimes, 1); | |
| 479 | } else { | ||
| 480 | // Only if this process is not the last one, act as a blocking call | ||
| 481 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 8 times.
|
15 | if (barrier->flags.lewi) { |
| 482 | 7 | sync_call_flags_t mpi_flags = (const sync_call_flags_t) { | |
| 483 | .is_dlb_barrier = true, | ||
| 484 | .is_blocking = true, | ||
| 485 | .is_collective = true, | ||
| 486 | .do_lewi = true, | ||
| 487 | }; | ||
| 488 | 7 | into_sync_call(mpi_flags); | |
| 489 | } | ||
| 490 | |||
| 491 | // Barrier | ||
| 492 | 15 | pthread_barrier_wait(&barrier->barrier); | |
| 493 | |||
| 494 | // Recover resources for those processes that simulated a blocking call | ||
| 495 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 8 times.
|
15 | if (barrier->flags.lewi) { |
| 496 | 7 | sync_call_flags_t mpi_flags = (const sync_call_flags_t) { | |
| 497 | .is_dlb_barrier = true, | ||
| 498 | .is_blocking = true, | ||
| 499 | .is_collective = true, | ||
| 500 | .do_lewi = true, | ||
| 501 | }; | ||
| 502 | 7 | out_of_sync_call(mpi_flags); | |
| 503 | } | ||
| 504 | } | ||
| 505 | |||
| 506 | 47 | unsigned int participants_left = DLB_ATOMIC_SUB_FETCH(&barrier->count, 1); | |
| 507 | 47 | bool last_out = participants_left == 0; | |
| 508 | |||
| 509 |
1/4✗ Branch 0 not taken.
✓ Branch 1 taken 47 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
47 | verbose(VB_BARRIER, "Leaving barrier %s%s", barrier->name, last_out ? " (last)" : ""); |
| 510 | |||
| 511 | /* WARNING: There may be a race condition with the 'count' value in | ||
| 512 | * consecutive barriers (A -> B), if one process increases the | ||
| 513 | * 'count' value after entering B, while another process still | ||
| 514 | * hasn't decreased it in B. Anyway, this is completely harmless | ||
| 515 | * since it only affects the verbose message. | ||
| 516 | */ | ||
| 517 | } | ||
| 518 | 47 | pthread_rwlock_unlock(&barrier->rwlock); | |
| 519 | } | ||
| 520 | |||
| 521 | 8 | void shmem_barrier__print_info(const char *shmem_key, int shmem_size_multiplier) { | |
| 522 | |||
| 523 | /* If the shmem is not opened, obtain a temporary fd */ | ||
| 524 | 8 | bool temporary_shmem = shm_handler == NULL; | |
| 525 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 7 times.
|
8 | if (temporary_shmem) { |
| 526 | 1 | shmem_barrier_ext__init(shmem_key, shmem_size_multiplier); | |
| 527 | } | ||
| 528 | |||
| 529 | /* Make a full copy of the shared memory */ | ||
| 530 | 8 | shdata_t *shdata_copy = malloc(shmem_barrier__size()); | |
| 531 | 8 | shmem_lock(shm_handler); | |
| 532 | { | ||
| 533 | 8 | memcpy(shdata_copy, shdata, shmem_barrier__size()); | |
| 534 | } | ||
| 535 | 8 | shmem_unlock(shm_handler); | |
| 536 | |||
| 537 | /* Close shmem if needed */ | ||
| 538 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 7 times.
|
8 | if (temporary_shmem) { |
| 539 | 1 | shmem_barrier_ext__finalize(); | |
| 540 | } | ||
| 541 | |||
| 542 | /* Initialize buffer */ | ||
| 543 | print_buffer_t buffer; | ||
| 544 | 8 | printbuffer_init(&buffer); | |
| 545 | |||
| 546 | /* Set up line buffer */ | ||
| 547 | enum { MAX_LINE_LEN = 128 }; | ||
| 548 | char line[MAX_LINE_LEN]; | ||
| 549 | |||
| 550 | 8 | int num_barriers = shdata_copy->num_barriers; | |
| 551 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 8 times.
|
16 | for (int i = 0; i < num_barriers; ++i) { |
| 552 | 8 | barrier_t *barrier = &shdata_copy->barriers[i]; | |
| 553 |
1/2✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 | if (barrier->flags.initialized) { |
| 554 | |||
| 555 | /* Append line to buffer */ | ||
| 556 | 8 | snprintf(line, MAX_LINE_LEN, | |
| 557 | " | %14s | %12u | %12u | %12u |", | ||
| 558 | 8 | barrier->name, barrier->participants, barrier->count, barrier->ntimes); | |
| 559 | 8 | printbuffer_append(&buffer, line); | |
| 560 | } | ||
| 561 | } | ||
| 562 | |||
| 563 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 1 times.
|
8 | if (buffer.addr[0] != '\0' ) { |
| 564 | 7 | info0("=== Barriers ===\n" | |
| 565 | " | Barrier Name | Participants | Num. blocked | Times compl. |\n" | ||
| 566 | "%s", buffer.addr); | ||
| 567 | } | ||
| 568 | 8 | printbuffer_destroy(&buffer); | |
| 569 | 8 | free(shdata_copy); | |
| 570 | 8 | } | |
| 571 | |||
| 572 | 8 | bool shmem_barrier__exists(void) { | |
| 573 | 8 | return shm_handler != NULL; | |
| 574 | } | ||
| 575 | |||
| 576 | 1 | int shmem_barrier__version(void) { | |
| 577 | 1 | return SHMEM_BARRIER_VERSION; | |
| 578 | } | ||
| 579 | |||
| 580 | 134 | size_t shmem_barrier__size(void) { | |
| 581 | // max_barriers contains a value once shmem is initialized, | ||
| 582 | // otherwise return default size | ||
| 583 | 134 | return sizeof(shdata_t) + sizeof(barrier_t) * ( | |
| 584 |
2/2✓ Branch 0 taken 133 times.
✓ Branch 1 taken 1 times.
|
134 | max_barriers > 0 ? max_barriers : mu_get_system_size()); |
| 585 | } | ||
| 586 |